Abstract

The salutary effect of formal education on health-risk behaviors and mortality is extensively documented: ceteris paribus, greater educational attainment leads to healthier lives and longevity. Even though the epidemiological evidence has strongly indicated formal education as a leading “social vaccine,” there is intermittent reporting of counter-education gradients for health-risk behavior and associated outcomes for certain populations during specific periods. How can education have both beneficial and harmful effects on health, and under which contexts do particular effects emerge? It is useful to conceptualize the influence of education as a process sensitive to the nature, timing of entry, and uniqueness of a new pleasurable and desirable lifestyle and/or product (such as smoking) with initially unclear health risks for populations. Developed herein is a hypothesis that the education gradient comprises multiple potent pathways (material, psychological, cognitive) by which health-risk behaviors are influenced, and that there can be circumstances under which pathways act in opposite directions or are differentially suppressed and enhanced. We propose the population education transition (PET) curve as a unifying functional form to predict shifting education gradients across the onset and course of a population’s exposure to new health risks and their associated consequences. Then, we estimate PET curves for cases with prior epidemiological evidence of heterogeneous education gradients with health-risk behaviors related to mass-produced cigarettes in China and the United States; saturated fats, sugar, and processed food diets in Latin America; and HIV infection in sub-Saharan Africa. Each offers speculation on interactions between environmental factors during population exposure and education pathways to health-risk behaviors that could be responsible for the temporal dynamics of PET curves. Past epidemiological studies reporting either negative or positive education gradients may not represent contradictory findings as much as come from analyses unintentionally limited to just one part of the PET process. Last, the PET curve formulation offers richer nuances about educational pathways, macro-historical population dynamics, and the fundamental cause of disease paradigm.

Introduction

The salutary effect of formal education on health-risk behaviors, disease, and mortality is extensively documented: ceteris paribus, greater educational attainment leads to healthier lives and longevity (e.g., Mirowsky and Ross 2003). This relationship between education and health-risk behaviors—the education gradient—is well established and referred to as a “social vaccine”: a social intervention that can protect individuals and elevate the health of populations (e.g., Beasley et al. 2008). For individuals, education is a key factor in moderating or avoiding health risks, such as tobacco use, alcohol abuse, illicit drug use, unhealthy diets, high-risk sexual practices, and sedentary lifestyles (e.g., Baker et al. 2011a; De Walque 2010; Gfroerer et al. 1997; Van Hook et al. 2013). Even modest amounts of formal schooling for mothers account for more effective use of medical resources for children’s health in low-income nations, and greater education attainment leads to better adherence to complicated medical treatment regimens for noncommunicable diseases and consistent use of preventive care in high-income nations (Case et al. 2005; Glied and Lleras-Muney 2008; Goldman and Lakdawalla 2005; Pamuk et al. 2011). Greater access and participation in formal education within populations is a likely contributor to the first demographic transition and is responsible for the age compression of mortality (Baker et al. 2011b; Brown et al. 2012; Lutz et al. 2001). Specific estimates have found that education reduces a variety of health-risk behaviors in high-income populations, and evidence suggests that disparities in education are associated with racial and ethnic health disparities net of other demographic characteristics, including wealth (Cutler and Lleras-Muney 2010; Hummer et al. 2004; Link and Phelan 1995).

At the same time intermittent reports suggest positive education gradients for health-risk behaviors and diseases for certain populations during specific periods. For example, studies have reported that greater educational attainment can be associated with HIV infection, greater tobacco use, cocaine use, and higher rates of overweight (Coombe and Kelly 2001; De Walque 2010; Fortson 2008; Fu et al. 2014; Hargreaves and Glynn 2002; Jeon et al. 2015; Leon et al. forthcoming; Miech 2008; Pi-Sunyer 1999; Smith et al. 2012). Correspondingly, an assessment of the education gradient for various diseases across birth cohorts indicates a temporal pattern of positive gradients among older cohorts and negative gradients among younger cohorts (Cutler and Lleras-Muney 2006, 2010). With no evidence to the contrary, assuming that reported shifting gradients are not a function of extensive measurement error or spurious associations, they represent a gap in the knowledge about education, health, and population trends. Namely, how can education have salutary as well as harmful gradients with health, and under which population contexts do particular gradients emerge? To address this puzzle, we offer four contributions to propose—albeit not definitively test—a unifying hypothesis about why and when education gradients shift.

First, we assess research on the different types of resources that education supplies an individual, establishing that education likely has various causal pathways to consequential health behavior. Further, we suggest that pathways can act in concert or disunity. If so, then the education gradient results from multiple potent pathways by which health behavior can be influenced; and although these pathways often operate in a similar salutary direction, populations can experience environmental conditions under which pathways act in opposite directions or are differentially suppressed and enhanced. Given this, the direction of the education gradient may follow a temporal pattern, which when plotted over the course of a population trend or transition that yields exposure to a new health risk suggests a unifying functional form to describe the change in the education gradient.

Second, we estimate PET curves to explore the utility of this formulation. The most relevant circumstances to investigate are during the introduction and increased exposure to new desirable, pleasurable, and/or social status–enhancing lifestyles and products in populations that eventually become identified as harmful to health, often resulting in epidemics, pandemics, and health care crises. We estimate five PET curves using data from recent studies in the United States, China, Latin America, and sub-Saharan Africa on drug use, diet, and sexual behavior reporting heterogeneous education-risk gradients across increasing exposure to mass-produced cigarettes; saturated fats, sugar, and processed food diets; and HIV/AIDS infection (De Walque 2010; Fu et al. 2014; Jeon et al. 2015; Leon et al. forthcoming).

Third, we suggest that the PET process is caused by macro-level events that occur over the course of a population’s exposure to a new health risk, influencing the strength and unity of education’s multiple pathways to behavior. As initial examples, we plot selected environmental events against the historical timing of some of the estimated PET curves and speculate on possible influences of each. A future test of the PET curve hypothesis will require detailed data on macro- and micro-level processes over long historical periods covering the introduction and subsequent increase of a new lifestyle and/or product with implications for health behavior across a range of risks and related diseases.

Last, we conclude by exploring ways by which the PET curve may further develop the fundamental cause of disease theory (FCT). Because education is a central component of the FCT, the PET curve formulation and its empirical demonstration squarely rest on the social shaping of health paradigm. Initially proposed to move the “modern school” of epidemiology beyond only proximate biological causes of disease, the FCT reasons that social factors, such as educational attainment, put people at more or less risk of more-proximate causes of disease and are thus responsible for health disparities within populations (Link and Phelan 1995, 1996, 2000). Therefore, theoretically, it is best to consider the FCT as the more general paradigm and the PET curve as a specified formulation of how this occurs in a complex fashion for one major social factor.

Education’s Multiple Pathways and the PET Curve Hypothesis

Educational attainment is a dynamic process that transforms individuals in multiple ways by providing them with an array of resources influencing health behavior and longevity (Baker 2014; Lutz and Samir 2011; Mirowsky and Ross 2003; Montez et al. 2011). The main pathways to behavior from increased education influencing later-life health include wealth and status effects, cognitive benefits, and psychological empowerment. Of course, many resources build off these main ones from education, and synergy among them can be robust. To simplify the ideas here, however, only the main pathways are considered. First, material wealth and social status are now so widely a function of earned educational credentials that they can be considered as part of education’s pathway to better health. Educational attainment is a necessary requirement for an increasing proportion of jobs in labor markets worldwide and determines earnings and occupational prestige (e.g., Bills 1987). Further, over the historical expansion of education, the independent impact of the individual’s educational attainment on economic and status attainment has come to outweigh much of the impact of the socioeconomic status (SES) of family of origin (e.g., Hout 1988, 2012).

In addition to wealth and social status, and cognitive resources from education have substantial influence on health behavior. Basic literacy and numeracy gained through schooling are necessary to recognize and accumulate information on health risks and preventative measures, yet education’s enhancement of cognition goes deeper than the capacity to obtain information and do fundamental mathematics and language skills (Smith-Greenaway 2013). Even a few years of schooling yield a greater propensity for more accurate decision-making and effective use of information to solve unique problems, and such thinking skills are twice as efficacious as just knowing facts about health (e.g., Baker et al. 2012; Duncan et al. 1996). For example, in field experiments in Ghana and Peru with subsistence-level farmers ranging from no formal education to secondary schooling, exposure to schooling enhanced cognitive functioning in terms of larger working memory, better planning skill, effective attention shifting, and accurate perception of the spread of an infectious disease (Peters et al. 2010). Evidence also suggests that learning activities in schooling assists in fuller development of neurocognitive capacity continuing through more advanced levels of education (Baker et al. 2015). Last, school-enhanced cognitive skill significantly mediates the association between educational attainment and health, and thus likely represents a pathway to effective health knowledge leading to attitude and behavior change among more educated populations (Cacioppo and Petty 1979; Herd 2010).

The psychological empowerment effects of education on individuals are often taken-for-granted outcomes of formal schooling, yet these also shape how education influences health decision-making among individuals, families, and communities (e.g., Caldwell 1993). Early comparative studies have shown that modest amounts of educational attainment lead to greater personal autonomy and sense of efficacy, as well as the inclination to “think for oneself” separate from traditional sources of authority (e.g., Inkeles 1996). In addition, education may enhance perception and enactment of ideations around fertility, family formation, and health decisions and behaviors (Thornton 2001). The clearest evidence of the psychological empowerment effects of formal schooling is on maternal and child health, as young mothers from traditional societies with modest amounts of formal schooling enact their health knowledge in seeking effective child health care (LeVine et al. 2004; Greenaway et al. 2012). An extensive review across a large number of studies consistently found that educationally derived empowerment to act and negotiate more independently among young women leads to more contraceptive use and better fertility health outcomes (Peters et al. 2014). Greater exposure to education increases mothers’ inclination to think through new problems, apply known facts more effectively, and use reasoning in self-advocacy with other institutions, thereby enhancing their access to, and enactment of, better health regimes through improved health knowledge (Glewwe 1999).

Therefore, education provides individuals with an array of resources—material and nonmaterial—and is a prominent component of the FCT of social shaping of health (Link and Phelan 1995). As diagrammed in Fig. 1, when multiple education-driven pathways and their possible interactions are working in a similar beneficial fashion, they are responsible for widely reported negative education gradients. Material and status resources can lead to access to better health care, cognitive resources can lead to better conceptualization of risks and problem-solving for prevention, and psychological resources can lead to empowerment to think and act effectively about risks. For example, an analysis across six British and American epidemiological data sets on health risks and preventive measures estimated that 20 % of the total beneficial education gradient is owed to material status resources; 30 %, to cognitive resources; and 12 %, to psychological resources (Cutler and Lleras-Muney 2010). It is likely that dynamic interactions among these three major education pathways produce additional effects.

Reports of heterogeneous education gradients in health-risk behavior, however, point to the possibility that pathways do not always act in concert relative to healthier behavior; and given that each is potentially a significant pathway, negative education gradients can be lessened and even reversed. For example, material status resources from education can purchase better health care and advice, but these resources can also facilitate greater risk-taking and consumption of potentially unhealthy substances and lifestyles; cognitive resources are essential for accurate decision-making about behavior, but in an environment with a lack of accurate health information, the cognitive analysis of the information often reveals no risk; and psychological resources can lead to prevention-taking, yet they also can empower greater risk-taking (Marmot 2004). Further, one pathway’s harmful potency at a particular point in time can suppress the beneficial effects of the other pathways.

The nature, timing of entry, and uniqueness of a risk within a population’s environment likely determine the congruence, disunity, or differential suppression of education’s multiple pathways to health behavior. The cases examined here illustrate this over the course of a population’s exposure to new desirable, pleasurable, or status-enhancing lifestyles and/or products with potential risks that are unclear or unknown for some time.

For example, consider tobacco consumption, a pleasurable stimulate that has been an ageless behavior albeit only among some strata of some human populations at certain periods. Introduction of new products (i.e., nicotine delivery mechanisms), however, rapidly exposed large portions of humans to smoking. The availability of inexpensive, mass-produced cigarettes early in the twentieth century increased the popularity and status enhancement from smoking. Therefore, more money for non-essential goods, early recognition of new status-enhancing trends, and the empowerment to try and incorporate new behaviors and products—all positively influenced by the material status and psychological resources from education—likely contributed to greater cigarette consumption by higher-educated individuals.

At the same time, a lack of information on potential risks of smoking essentially neutralized cognitive resources from education for reasoning about this health-risk behavior. Eventually new authoritative information entered the environment that made education-enhanced cognitive resources more prominent. After educated individuals applied better reasoning skills to accurate information yielding informed judgments of risk, they were more likely to be among the first to reduce and stop smoking, now marshaling the same psychological and material status resources for risk prevention guided by better risk judgments. Subsequently, the initial positive education-smoking gradient began to decline and ultimately turned negative (De Walque 2010; Link 2008).

A similar process can occur with other desirable and pleasurable lifestyle changes in populations: increased access to sedentary occupations; new desirable products, such as access to kinds of foods; new forms of illegal drugs; and, even an interaction of a new virus with traditional, desirable, and status-oriented sexual practices. The PET curve formulation is offered as a unifying framework for all such cases and, as will be shown, is compatible with earlier epidemiological findings of curving education gradients across a number of health-risk behaviors and diseases (Cutler and Lleras-Muney 2006, 2010). Figure 2 displays a hypothetical PET curve of time-varying coefficients of the association between education and odds of health-risk behavior over the course of a population’s exposure to a new lifestyle and/or product. There are two hypothesized parts to how this unfolds. First, the PET curve hypothesizes a predictable temporal shift in the direction of the education–new risk gradient. As described earlier, a positive gradient can occur with some of the education pathways early in the population’s exposure to a new pleasurable lifestyle/product that includes health risks that are unclear or unknown. This continues until an environmental change activates all of education’s pathways into a congruent causal force, thus decreasing odds of unhealthy behavior among the more educated (Smith et al. 2015).

The second part of the PET curve hypothesis indicates that macro-level environmental influences can shape the direction, consistency, and relative strength of the multiple pathways of education relative to the health risk. For example, shown in Fig. 2 is the entrance into the population’s environment of previously unknown health-risk information about the new lifestyle or product as a potential cause of the onset of a decline in the positive education gradient. And as we speculate later, there are likely several types of environmental factors that could drive the PET process. Some—such as extensive marketing of products or lifestyles, political suppression of accurate public health information, and inept public health campaigns—can lead to disunity and suppression of pathways. Others—such as new research findings, sudden availability of accurate risk information, and initiation of effective public health campaigns—can lead to greater congruence among education’s pathways. Broader educational, economic, and political institutional change could also influence the process.

Estimated PET Curves

Methods

Five PET curves are estimated for different populations and periods from approximately the onset of exposure to the following: (1) mass-produced cigarettes in United States (1940–2000) and (2) in China (1940–2010); (3) access to saturated fats, sugar, and processed food diets in nine Latin American and Caribbean countries (1990–2010); (4) sexual exposure to HIV infection in Uganda and (5) in Tanzania (1980–2010).

Prior research on each case reported heterogeneous education-risk gradients over the course of exposure. Estimations of the PET curves use the same data, sample, measurement of education, definition of the health risk, and conditioning variables as those analyzed in each prior study; details of these are summarized in Table 1, and estimated coefficients are displayed in Tables S1–S5 in Online Resource 1. Because of historical and regional variations in expansion and access to mass education, levels of high and low educational attainment differ across the cases. Additionally, some macro-level events occurring over the course of a population’s exposure to the new lifestyle and/or product that could have influenced the strength and unity of education’s multiple pathways to behavior are plotted against their corresponding PET curves. Last, to contextualize each PET curve historically, we include an indicator of the population’s access to the product or exposure to the risk over the analyzed period.

Analytic Procedure for Estimating PET Curves

Consumption of Mass-Produced Cigarettes in the United States

Estimated coefficients for the case of mass-produced cigarettes in United States were extracted from De Walque’s (2010) analyses of the association between education and smoking behavior for 10-year birth cohorts between 1910 and 1979. Using repeated cross-sectional versions of the National Health Interview Survey from 1978 to 2000, De Walque constructed detailed smoking histories at the time of the interview, and then estimated a two-period panel regression with individual fixed effects to control for the influence of nonobservable time-invariant factors. These two periods are smoking at age 17 and smoking at ages 25 to 60. The model estimated by De Walque is
Yti=βXti+ηi+γt+εti,
1

where t indicates times at age 17 or at ages 25–60; Yti is a binary variable for whether the individual smokes at age t; Xti is a binary variable for whether the individual is a college graduate at time t; ηi is a time-invariant individual specific fixed effect; γt is an age effect; and εti is the error term. Then this model is estimated for different birth cohorts, and the education-smoking gradient for each cohort is the estimate of β. For the purpose here, coefficients for males aged 30 and older are extracted and plotted by decade over the period 1949–1999.

Consumption of Mass-Produced Cigarettes in China

The 2010 Global Adult Tobacco Survey (GATS), a nationally representative household survey conducted in 28 provinces in China, yielded a sample of 3,500 regular daily smokers and 9,344 nonsmokers older than age 15 (Fu et al. 2014). Logit regressions for individual birth cohorts of males and females aged 15 and older from 1940 to 2010 estimate the education-smoking gradients across time as follows:
Yi=β0+β1Xi+β2Ci+εi,
2

where Yi is a binary variable for each individual i, identifiying the smoking status at each birth cohort; Xi is a binary variable differentiating complete secondary or less versus incomplete college or more; Ci is a group of control variables including gender, area of residence, region, and a family wealth index (composite score of 11 facilities and appliances, such as flush toilets, televisions, and cars); and εi is the error term. The education-smoking gradient for each birth cohort is the estimate of β1, and these are plotted by decade over the period.

Consumption of Saturated Fats, Sugar, and Processed Food Diets in Latin America/Caribbean

The Demographic and Health Survey (DHS) has been administered at different years in eight Latin American countries: Bolivia (1994, 1999, 2003, 2008); Brazil (1996); Colombia (1995, 2005, 2010); the Dominican Republic (1991, 1996); Guatemala (1995, 1999); Honduras (2006); Nicaragua (1998, 2001); and Peru (1992, 1996, 2002, 2008). The DHS has also been administered in the bordering Caribbean country of Haiti (1995, 2000, 2006). Because DHS data are not available for any nation over the entire course of growing exposure to saturated fats, sugar, and processed food diets, the authors of the original study assigned each nation’s DHS administration year to quintiles of access to these diets (Jeon et al. 2015). This was determined by national estimates of the total energy expenditure per day per capita, the energy derived from each dietary source, and the consumed weight of each dietary source from the Food and Agriculture Organization—a scale referred to as the nutritional transition (NT)—wherein countries further along in the NT consume higher amounts of saturated fats, sugar, and processed food diets (Popkin and Gordon-Larson 2004). Also, as is common in obesity research, Jeon et al. (2015) limited the DHS samples to mothers aged 15–49 because past research has suggested that women of reproductive age tend to be more sensitive to body shape, and results indicate a strong relationship between mother’s education and overweight gradients, and a reduction in the age-specific confounding effect present when including younger and older women. Merging all nation’s DHS units yields an analysis sample of 143,258 mothers aged 15–49. Multilevel logit regressions were estimated at each quintile of the NT:
Atthe individual level:Yij=β0j+β1jXij+β2jCIij+εij,andatthe population level:β0j=γ00+γ01CPj+ν0jβ1j=γ10+γ11CPj+ν1jβ2j=γ20,
3

where Yij is a binary variable for mother’s overweight (body mass index (BMI) > 25.0 kg/m2); Xij is a binary variable differentiating incomplete secondary or less versus complete secondary education or more; CIij is a group of control variables at the individual level, including age, area of residence, pregnancy, and a family wealth index; CPj is a group of control variables at the population level, including log of GDP per capita and percentage of the population living in an urban area; and εij, ν0j, and ν1j are the error terms at the individual and population levels, respectively. The education-overweight gradient for each quintile is the estimate of β1, and these are plotted over the NT quintiles of population access to saturated fats, sugar, and processed-food diets.

Exposure to HIV in Sub-Saharan Africa

The DHS-2011 yields an analyzable two-stage stratified random sample of 21,359 and 17,702 men and women aged 15–59 from Uganda and Tanzania, respectively. Unlike higher refusal rates of the biomarker in previous DHS administrations, only 1 % in each country sample refused to participate in a dried blood spot test on a filter paper card to determine HIV infection status. Using responses about sexual history, subjects were divided into year of sexual initiation cohorts from 1980 to 2010, and then a cohort logit regression was estimated:
Yi=β0+β1Xi+β2Zi+β3Xi×Zi+β4Ci+εi,
4

where Yi is a binary variable for each individual i, coded as 1 if the result of the blood test was positive and 0 otherwise. Xi is a binary variable differentiating complete primary or more versus incomplete primary or less; Zi is a set of binary variables indicating year of sexual initiation cohorts; Xi × Zi is the interaction term between education level and the specified cohorts; Ci is a group of control variables including gender, marital status, age, area of residence, risky sexual behavior (defined as two or more sexual partners other than the current partner), have ever had sex, HIV/AIDS knowledge (measured by three questions of basic ways to avoid HIV/AIDS infection), wealth index (composite score of 10 assets or services in the house), and geographical region. The education-infection gradient for each year of sexual initiation cohort from 1980 to 2010 will be the estimate of β3, which is the set of estimated coefficients for the interaction term Xi × Zi. Finally, estimated coefficients for both countries are plotted separately over the period.

Estimated PET Curves

Education-Smoking PET Curves

The tobacco-related disease pandemic, resulting chiefly from mass-produced affordable cigarettes, is responsible for an estimated 100 million deaths worldwide over the course of the twentieth century: tobacco consumption is a risk factor for six of the eight leading causes of death (World Health Organization (WHO) 2011). With a growing tobacco industry and changing social mores, the overall supply and annual per capita cigarette consumption of American adults increased steadily to more than 4,500 cigarettes by mid-century (De Walque 2010).

Panel a in Fig. 3 displays the estimated PET curve for educational attainment and log odds for smoking among U.S. males over the rise in access to mass-produced cigarettes. By 1949, Americans aged 30 and older with a college degree were more likely to smoke daily. Starting in the late 1950s, the positive odds of cigarette smoking among those with some college or more declined, reaching a negative odds relative to the less-educated by the late 1970s; the social vaccine effect increased in magnitude until the early 1990s. By 1999, the probability of a college-educated individual smoking (14.2 %) was less than one-half that of an individual without a high school diploma (29.2 %). The first of three vertical lines on the curve indicates that 1950 begins a series of influential American and British epidemiological publications reporting an association between cigarette smoking and lung cancer (Alberg et al. 2014; Doll and Hill 1950). The second vertical line marks the 1964 U.S. Surgeon General’s report on the health risks of smoking, the first of a number of influential public health statements. The third vertical line indicates the change in health warnings on cigarette packaging mandated by 1971 legislation. Although a flattening of the positive gradient occurred before these events, they coincide with the transition toward a negative odds ratio by the late 1970s.

Panel b in Fig. 3 displays the estimated PET curve for educational attainment and log odds for smoking among Chinese adults from 1940 to 2010. With educational, economic, and political change occurring over the period, the curve for China is more complex, and the education gradients are more pronounced than in the United States. In the 1940s, college-educated individuals in China were more likely to smoke (probability = 56.1 %) relative to those with less than a college education (43.9 %). Through the 1950s, greater levels of education were associated with an increased likelihood of smoking. The positive gradient decreased into the mid- to late 1950s. After this period, at least some college education acted as a social vaccine: those with a college degree were less likely to smoke in the 1960s (46.9 %) than those with less than college (53.1 %). Interestingly and as discussed later, in contrast to the United States, the ameliorating effect of greater levels of education decelerated as China entered the twenty-first century.

Education-Overweight PET Curve

Overweight leading to obesity and numerous accompanying health hazards has become a global epidemic reaching many middle- and low-income countries (Caballero 2005). In the Latin America/Caribbean region, the prevalence of overweight and obesity is often equal or higher than that in wealthier national populations, and all the countries analyzed here except for Haiti currently have an overweight rate higher than 50 % (Van Hook et al. 2013; WHO 2013). Beginning with diets dependent on cereals and low in fats and sugar, with considerable risk of nutritional deficiencies, populations can move into the stages of the NT caused by greater access to the “Western diet”—saturated fats, sugar, and processed food, but low in fiber—leading to changes in average stature, body composition, and morbidity (Popkin and Gordon-Larson 2004).

Figure 4 shows the estimated PET curve for educational attainment and log odds of overweight among mothers aged 15–49 as the mean proportional caloric intake of animal fats and sweeteners doubled in this region’s population from 15 % to 30 %. As discussed later, this PET curve generally confirms to the hypothesized pattern but lacks an increasing positive gradient at early stages and displays a gradient becoming gradually more negative in later stages.

Education-HIV Infection PET Curves

Populations in sub-Saharan Africa continue to face high risk of exposure to the HIV and are the epicenter of a world HIV/AIDS epidemic estimated to have infected 78 million individuals, causing approximately 39 million deaths. In this region, nearly 20 % of adults are HIV-positive, accounting for 71 % of the people living with HIV worldwide (WHO 2015). Although prevalence rates vary across national populations in the region, rates increased significantly from at least 1980 until the mid-1990s.

As shown in Fig. 5, the Ugandan curve conforms to the expected rise of a positive education gradient through the earliest known population exposure, which transitioned to a negative gradient in the early 1990s as the prevalence rate in the country spiked. The Tanzanian curve shows a different timing. Similar to Uganda, Tanzanian adults with at least a primary school education (i.e., higher than average education for this population and period) were at a substantially greater risk of infection early in the epidemic. However, instead of transitioning to a social vaccine in the early 1990s, the relationship between education and risk was neutral (hovering around an odds of risk of 0) until 2005, when a substantial negative gradient began.

Published reports and assessments of the governmental and public health responses in the region suggest that these two national populations went through three differing periods in the quality of information and public health environment during the pandemic. Vertical lines on each curve demarcate these periods and their durations (De Walque 2007; Epstein 2007; Grmek 1990; Leon et al. forthcoming). In both countries, the progression to an environment with some accurate information and partial public health mobilization coincides with the peak positive gradient and the beginning of its decline. Uganda’s faster transition to dissemination of accurate information and full mobilization of public health actions is associated with a steady decline in the positive gradient and increasing negative gradient up to 2005. Tanzania’s slower transition is associated with a flatter negative gradient until 2005.

PET Curve as a Unifying Form and Possible Causes

The estimated curves suggest a general functional form from which to predict the temporal course of education gradients over exposure to new lifestyles and/or products with health implications. They also indicate that the PET hypothesis is feasible and warrants future testing. In each case, net of other characteristics of the individual, more educational attainment was associated with greater odds of behaviors leading to health risks early in a population’s exposure; as exposure increased, the positive gradient decreased and eventually turned negative. Thus, past epidemiological studies reporting either negative or positive education gradients may not represent contradictory findings as much as come from analyses unintentionally limited to just one part of the PET process. In comparison with the more frequently cited negative gradients, positive education gradients are not anomalous because they could be routine occurrences over the temporal course of a population’s exposure to new lifestyles and/or products with health implications.

Whether the PET curve is a suitable functional form for a wide variety of health risks awaits future research. In each case, though, it is plausible that at the onset of population exposure, more-educated individuals’ relative material resources, status, and psychological propensity to empowerment played a role in greater consumption of cigarettes, saturated fats, sugar, processed food, and sex with multiple partners. Then as health risks became apparent, educated individuals, on average, reasoned more accurately about risks and ceased these behaviors. For example, although the data here could not estimate the PET curves from the absolute earliest onset of population exposure, prior research on the onset of each risk documents initial growing positive gradients. During early access to mass-produced cigarettes (circa 1900 to the 1930s) in the United States, smoking was portrayed as a desirable social status marker through media and aggressive tobacco industry marketing; therefore, educated Americans during this time used resources to consume cigarettes (Brandt 2007). A similar process likely occurred in China. Prior to the Communist regime takeover in 1949, production of cigarettes was limited, making them expensive and a status symbol for elites, including the more-educated (Fu et al. 2014). Corresponding scenarios for growing positive education gradients are reported in national populations during initial access to saturated fats, sugar, and processed food (Martorell et al. 2000).

The HIV case is also consistent with this but with the exception that a virus was introduced into preexisting practices of long-term concurrency of multiple sexual partners and transactional sex consumption by more-educated and hence higher-status and materially resourced males. A tradition of long-term concurrency of partners in intimate relationships might have meant that the virus rapidly infected networks of sub-Saharan African adults without engagement in risky sexual practices identified in the Western experience with the disease, and educational resources could contribute to access and maintenance of such long-term concurrent intimate networks (Cutler and Lleras-Muney 2010; Epstein 2007; Morris and Kretzschmar 1997). For example, ethnographic data have demonstrated this dynamic behind an initial positive gradient in which educated males were early transmitters of the virus (Swidler and Watkins 2007; Wamoyi et al. 2011). At the beginning of population exposure, each curve begins with substantial positive log odds—and in the Ugandan HIV case, an increasing positive education gradient.

In each PET curve case, as population exposure increased, the positive gradient decreased, sometimes sharply, and eventually moved to a negative gradient. As suggested earlier, this could be a function of an environmental change that enables the average superior cognitive and decision-making skills of the more-educated to play a positive role in enacting preventative behavior related to the new risk (Baker et al. 2011a, 2015). Additionally, as a fuller understanding of the risk emerges, more-educated individuals can marshal material and psychological resources from education to reduce their risk; thus all the resources would be employed in a salutary fashion.

The second part of the PET hypothesis predicts a causal link between the dynamics of PET curves and environmental changes in factors such as public health prevention campaigns, scientific research, and quality of information disseminated in mass media. Although this complex hypothesis cannot be tested here, the timing of changes in the information environment, and other institutional factors for the tobacco and HIV cases, offer some preliminary speculation on its feasibility.

The cognitive resources of education could have been activated with new, accurate, and persuasive information on the health risk of cigarette smoking, which initially was thought of as a pleasurable and status-enhancing activity (e.g., Baker et al. 2015; Peters and Büchel 2011; Peters et al. 2006). Although German epidemiologists published the earliest scientific evidence of tobacco-related chronic diseases in the late 1930s, a stigma from connection to the Nazi government’s anti-smoking campaign kept the findings from being widely reported elsewhere (e.g., Müller 1940). It was not until the 1950s that American and British epidemiological studies, reporting results that heavy smokers of cigarettes were 50 times as likely as nonsmokers to contract lung cancer, begun to be disseminated through the mass media. Starting in 1964, a series of governmental public health reports and legislation appeared (Brandt 2007). The change in the informational environment of the U.S. population occurred as the magnitude of the positive gradient declined, while changes in health warnings on cigarette packaging mandated by 1971 legislation occurred several years prior to the shift to a negative gradient. Indeed, Link’s (2008) reassessment of the historical timing of information about cancer, public attitudes, and emergence of a negative education-smoking gradient in the United States foreshadows the concept here of the historical macro-contexts of education’s role in shaping a number of health-risk behaviors.

The Chinese smoking PET curve is likely a function of several intertwined public health, economic, and political institutional factors. By the 1940s and the nationalization of the cigarette industry, the new Communist government mass-produced inexpensive cigarettes and did not undertake anti-smoking campaigns during the period of scientific and public health warnings in the West. Smoking rose rapidly across the population, which at this time was mostly unschooled or minimally educated. The relatively small number of educated individuals likely had the resources to access the scientific and public health information emerging in the West in the 1960s, and this coincides with the decline in the positive education gradient. The timing of the curve also suggests that with greater economic growth and more youth with more education and cosmopolitan experiences in the 1980s, the negative gradient flattened out and may have decreased as cigarette smoking perhaps became a “new” status symbol of an economically developing China.

Inaccurate public health messages and government reluctance and incompetence in responding to the HIV/AIDS crisis in many sub-Saharan African nations has been documented; tragic in its human toll, this occurrence adds further insight into the underlying PET process. The timing of the PET curves is associated with changes in the quality of information and public health campaigns in each country. Uganda’s generally acknowledged faster and higher-quality public health response coincides with a consistent increase in the negative education gradient. Tanzania’s slower response is associated with the neutral relationship between education and risk, with a substantial negative gradient only beginning to emerge in the mid-2000s (e.g., De Walque 2007; Epstein 2007; Grmek 1990).

Because national populations in Latin America/Caribbean were exposed to rising access to saturated fats, sugar, and processed food at different historical points, the PET curve is plotted against NT quintile stages, with the earliest DHS data from Peru in 1992 to the most recent from Columbia in 2010. An environmental factor that could have started the change in the education gradient specific to Latin America/Caribbean is not yet clear because to our knowledge, no systematic research has studied the public health response to diet change in these countries. These data, however, were collected well after considerable research and public health information came to light about the risk of overweight from the Western diet in other populations, which could have served as some environmental influence on the PET process for any nation further into the NT (McLaren 2007; Mokdad et al. 1999).

This case provides further insight into the temporal process of PET curves. In the initial study, Jeon et al. (2015) plotted the probability of mothers being overweight by level of education attainment separately for populations during low, medium, and high access to saturated fats, sugar, and processed food in the national food supply. They found that the specific education level at which the gradient turned negative declined over the NT from 10 to 7 to 3 years, respectively. Thus, as a population gains more experience with a new health risk, perhaps less educational resources are required to initiate healthier behavior. The negative education gradient may have a contagion effect with more people, probably at first with the more-educated, exhibiting health alternatives that could be modeled by the less-educated. The same effect could be true for an increasing positive gradient (Marmot 2004).

Conclusion

The empirical cases illustrate that a PET curve could be a plausible representation of shifting education gradients as a population experiences onset and increasing exposure to a new lifestyle and/or product with health implications, particularly when the factor has some desirable, pleasurable, and/or status-enhancement qualities. If future research supports this, there are several theoretical and policy implications.

As noted earlier, the PET curve can be thought of as a recurring, complex case of the general paradigm of the FCT about the social shaping of health disparities. Seen this way, the PET curve advances the social shaping paradigm and points to future research. First, formal education through schooling is arguably becoming the most effectual shaping factor among all those hypothesized by the FCT. At the same time, as a macro-social institution, education is central to contemporary society in both its demographic and cultural rise over the past 150 years (Baker 2014). Therefore, the PET curve process expands the FCT to include larger institutional processes, including the expansion of education supply, public health policy, technological innovation, political actions, and the marketplace. Second and related, the PET curve demonstrates a confluence of institutional factors, individual resources, human agency, and health behaviors that is new within the FCT paradigm. Third, because education provides individuals with a range of resources that can influence their health, the PET curve broadens consideration of causal micro-mechanisms behind the general FCT process. For example, the particularly potent cognitive resources from education are likely a result of neurocognitive development formed through a genetic-schooling environment synthesis, thus suggesting an avenue for social shaping research to consider the dynamic genetic-social environmental processes in health (e.g., Baker et al. 2012, 2015). Fourth, implications for population health from demographic change in the social factors themselves were less a part of the original formation of the FCT, and the PET curve brings focus to this. For instance, the predictable increase in formal education attainment across future birth cohorts worldwide has significant consequences for the social shaping of coming global population health trends, such as the epidemiological transition to greater chronic disease burden in populations (e.g. Hummer and Lariscy 2011; Smith et al. 2015). Related, educational expansion also needs to be incorporated into testing of the FCT in terms of persistence health disparities as a function of a steady stream of new emergent major causes of mortality (Miech et al. 2011). Expanding population levels of education is likely a cause of both diminishing disparities among some existing causes and the rise of disparities among new causes. Last, although its originators did not intend it, the FCT can be too narrowly interpreted as primarily a one-way linear resource model: more social resources (material and nonmaterial) decrease health-risk behavior and its outcomes. By explicitly predicting social shaping by education that includes both negative and positive gradients, the PET curve formulation advances the complexity, and perhaps accuracy, of the FCT paradigm.

In addition to its support and extension of the FCT, the PET curve formulation could assist in organizing reports of contrasting education gradients and thereby reduce confusion in research literature on some health risks. For example, early in the sub-Saharan HIV pandemic, epidemiological studies identified educational attainment as a risk factor, whereas later research argued that education was a social vaccine, leading some to question the overall strength of the association (e.g., Fortson 2008). Anticipation of a PET curve could motivate greater sensitivity to environmental changes over a population’s exposure to a risk in analyzing and interpreting education gradients and improve epidemiological projections.

The PET curve also suggests that education gradients are not weak just because they can change direction over time; the slope of the gradient on both sides of the curve illustrates this. Further, these cases suggest that the PET curves can reveal the speed with which education gradients change and how this might vary across populations and risks. For instance, it took more than three decades for a substantial negative education HIV gradient to occur in Tanzania, compared with a little more than a decade in Uganda. The reasons for such differences are open to speculation and future research, but the conclusion is that the timing of the PET curve can be a useful comparative indictor.

All the education gradients here eventually shifted to a negative gradient, which is in line with summary assessments of the salutary role of education in health disparities and mortality, and with the FCT in general (e.g., Mirowsky and Ross 2003; Montez et al. 2011; Smith et al. 2015). In theory, however, there could be trends during which the PET curve is first negative and then positive—or after turning negative reverses to positive, as may be occurring for consumption of cigarettes among the younger birth cohort in a more cosmopolitan China. This possibility suggests that future research on education gradients should be continued as long as the health risk exists in the population. In addition, although the health risks and education gradients here were examined in isolation from one another, in reality they can overlap. Smith et al. (2015), for example, demonstrated that PET curves and multiple health risks can interact to influence a population’s overall disease burden.

A full test of the PET curve hypothesis will require extensive data of two types. Four of the five PET curves (with the exception of mass-produced cigarettes in the United States) use cross-sectional regressions across birth cohorts to estimate gradients over specific periods. Selection effects due to differential mortality and less than optimal information on the timing of risk-taking are a limitation of this cross-section cohort approach. Therefore, a more rigorous estimation should employ panel data where it is possible to construct specific individual’s histories of health behavior and risk-taking. In addition, comprehensive data on environmental-institutional factors and matching microdata on the hypothesized pathways from education will be required to test the dynamics driving PET curves.

Last, understanding causal factors behind PET curves could be of assistance to the planning of public health responses to new health risks. Beyond new health risks, the PET curve process could inform related policy about population health in general because, for example, effective use by patients of treatments involving more technological innovation also has a large negative education gradient (Glied and Lleras-Muney 2008). Because education can eventually be a strong protective factor, policies and interventions to shorten the likely initial positive gradient and the period to the start of a negative gradient should be developed.

Acknowledgments

The authors thank Mark Hayward, Robert Hummer, Wolfgang Lutz, Jennifer Montez, Emily Smith-Greenaway, Jennifer Van Hook, and three anonymous reviewers for their helpful comments on earlier versions of the article. This research was partially funded by a National Research Foundation of Korea Grant (NRF-2016S1A3A2924944) to H. Jeon. With regard to affiliations for D. Salinas and W. Smith, the content of this article is solely that of the authors and does not represent an official position of the OECD or UNESCO.

References

References
Alberg, A. J., Shopland, D. R., & Cummings, K. M. (
2014
).
The 2014 Surgeon General’s report: Commemorating the 50th anniversary of the 1964 Report of the Advisory Committee to the US Surgeon General and updating the evidence on the health consequences of cigarette smoking
.
American Journal of Epidemiology
,
179
,
403
412
. 10.1093/aje/kwt335
Baker, D. P. (
2014
).
The schooled society: The educational transformation of global culture
.
Stanford, CA
:
Stanford University Press
.
Baker, D. P., Eslinger, P. J., Benavides, M., Peters, E., Dieckmann, N. F., & Leon, J. (
2015
).
The cognitive impact of the education revolution: A possible cause of the Flynn Effect on population IQ
.
Intelligence
,
49
,
144
158
.
Baker, D. P., Leon, J., & Collins, J. M. (
2011
).
Facts, attitudes, and health reasoning about HIV and AIDS: Explaining the education effect on condom use among adults in sub-Saharan Africa
.
AIDS and Behavior
,
15
,
1319
1327
. 10.1007/s10461-010-9717-9
Baker, D. P., Leon, J., Greenway, E. S., Collins, J., & Movit, M. (
2011
).
The education effect on population health: A reassessment
.
Population and Development Review
,
37
,
307
332
. 10.1111/j.1728-4457.2011.00412.x
Baker, D. P., Salinas, D., & Eslinger, P. (
2012
).
An envisioned bridge: Schooling as a neurocognitive developmental institution
.
Developmental Cognitive Neuroscience
,
2
,
S6
S17
.
Beasley, M., Valerio, A., & Bundy, D. (
2008
).
A sourcebook of HIV/AIDS prevention programs
.
Washington, DC
:
World Bank
.
Bills, D. B. (
1987
).
Costs, commitment, and rewards: Factors influencing the design and implementation of internal labor markets
.
Administrative Science Quarterly
,
32
,
202
221
. 10.2307/2393126
Brandt, A. M. (
2007
).
The cigarette century: The rise, fall and deadly persistence of the product that defined America
.
New York, NY
:
Basic Books
.
Brown, D. C., Hayward, M. D., Montez, J. K., Hummer, R. A., Chiu, C. T., & Hidajat, M. M. (
2012
).
The significance of education for mortality compression in the United States
.
Demography
,
49
,
819
840
. 10.1007/s13524-012-0104-1
Caballero, B. (
2005
).
The nutrition transition: Global trends in diet and disease
. In Shils, M. E., Sheke, M., & Ross, A. (Eds.),
Modern nutrition in health and disease
(pp.
1717
1722
).
Philadelphia, PA
:
Lippincott Williams & Wilkins
.
Cacioppo, J. T., & Petty, R. E. (
1979
).
Effects of message repetition and position on cognitive response, recall and persuasion
.
Journal of Personality and Social Psychology
,
27
,
97
109
. 10.1037/0022-3514.37.1.97
Caldwell, J. C. (
1993
).
Health transition: The cultural, social and behavioural determinants of health in the Third World
.
Social Science & Medicine
,
36
,
125
135
. 10.1016/0277-9536(93)90204-H
Case, A., Fertig, A., & Paxson, C. (
2005
).
The lasting impact of childhood health and circumstance
.
Journal of Health Economics
,
24
,
365
389
. 10.1016/j.jhealeco.2004.09.008
Coombe, C., & Kelly, M. J. (
2001
).
Education as a vehicle for combating HIV/AIDS
.
Prospects
,
31
,
435
445
. 10.1007/BF03220082
Creek, L., Capehart, T., & Grise, V. (
1994
).
U.S. tobacco statistics, 1935–92
(ERS Statistical Bulletin No. 869).
Washington, DC
:
Economic Research Service
.
Cutler, D. M., & Lleras-Muney, A. (
2006
).
Education and health: Evaluating theories and evidence
(National Bureau of Economic Research Working Paper No. 12352).
Cambridge, MA
:
National Bureau of Economic Research
. Retrieved from http://www.nber.org/papers/w12352.pdf
Cutler, D. M., & Lleras-Muney, A. (
2010
).
Understanding differences in health behaviors by education
.
Journal of Health Economics
,
29
,
1
28
. 10.1016/j.jhealeco.2009.10.003
De Walque, D. (
2007
).
How does the impact of an HIV/AIDS information campaign vary with educational attainment? Evidence from rural Uganda
.
Journal of Development Economics
,
84
,
686
714
. 10.1016/j.jdeveco.2006.12.003
De Walque, D. (
2010
).
Education, information and smoking decisions: Evidence from smoking histories in the United States, 1940–2000
.
Journal of Human Resources
,
45
,
682
717
. 10.1353/jhr.2010.0009
Doll, R., & Hill, A. B. (
1950
).
Smoking and carcinoma of the lung: Preliminary report
.
British Medical Journal
,
2
,
739
748
. 10.1136/bmj.2.4682.739
Duncan, J., Emslie, H., Williams, P., Johnson, R., & Freer, C. (
1996
).
Intelligence and the frontal lobe: The organization of goal-directed behavior
.
Cognitive Psychology
,
30
,
257
303
.
Epstein, H. (
2007
).
The invisible cure: Africa, the West, and the fight against AIDS
.
New York, NY
:
Farrar, Straus, and Giroux
.
China statistical yearbook
. (
1940
).
Rome, Italy
:
FAO
.
Food and Agriculture Organization (FAO) of the United Nations
. (
2013
).
FAOSTAT–Food Balance Sheets
[Data files].
Rome, Italy
:
FAO
. Retrieved from http://www.fao.org/faostat/en/#data/FBS
Fortson, J. G. (
2008
).
The gradient in sub-Saharan Africa: Socioeconomic status and HIV/AIDS
.
Demography
,
45
,
303
322
. 10.1353/dem.0.0006
Fu, T., Smith, W., Anderson, E., & Baker, D. (
2014
).
A cohort analysis of the effect of educational attainment on smoking behaviors in China
. Paper presented at the annual meeting of the Population Association of America,
Boston, MA
.
Gfroerer, J. C., Greenblatt, J. C., & Wright, D. A. (
1997
).
Substance use in the US college-age population: Differences according to educational status and living arrangement
.
American Journal of Public Health
,
87
,
62
65
. 10.2105/AJPH.87.1.62
Glewwe, P. (
1999
).
Why does mother’s schooling raise child health in developing countries? Evidence from Morocco
.
Journal of Human Resources
,
34
,
124
159
. 10.2307/146305
Glied, S., & Lleras-Muney, A. (
2008
).
Technological innovation and inequality in health
.
Demography
,
45
,
741
761
. 10.1353/dem.0.0017
Goldman, D. P., & Lakdawalla, D. N. (
2005
).
A theory of health disparities and medical technology
.
Contributions in Economic Analysis & Policy
,
4
,
1395
. doi:10.2202/1538-0645.1395
Greenaway, E. S., Leon, J., & Baker, D. P. (
2012
).
Understanding the association between maternal education and use of health services in Ghana: Exploring the role of health knowledge
.
Journal of Biosocial Science
,
44
,
733
747
. 10.1017/S0021932012000041
Grmek, M. D. (
1990
).
History of AIDS: Emergence and origin of a modern pandemic
.
Princeton, NJ
:
Princeton University Press
.
Hargreaves, J. R., & Glynn, J. R. (
2002
).
Educational attainment and HIV-1 infection in developing countries: A systematic review
.
Tropical Medicine & International Health
,
7
,
489
498
. 10.1046/j.1365-3156.2002.00889.x
Herd, P. (
2010
).
Education and health in late-life among high school graduates cognitive versus psychological aspects of human capital
.
Journal of Health and Social Behavior
,
51
,
478
496
. 10.1177/0022146510386796
Hout, M. (
1988
).
More universalism, less structural mobility: The American occupational structure in the 1980s
.
American Journal of Sociology
,
93
,
1358
1400
. 10.1086/228904
Hout, M. (
2012
).
Social and economic returns to college education in the United States
.
Annual Review of Sociology
,
38
,
379
400
. 10.1146/annurev.soc.012809.102503
Hummer, R. A., Benjamins, M. R., & Rogers, R. G. (
2004
).
Racial and ethnic disparities in health and mortality among the U.S. elderly population
. In Anderson, N. B., Bulatao, R. A., & Cohen, B. (Eds.),
Critical perspectives on racial and ethnic differences in health in late life
(pp.
53
94
).
Washington, DC
:
National Academies Press
.
Hummer, R. A., & Lariscy, J. T. (
2011
).
Educational attainment and adult mortality
. In Rogers, R. G. & Crimmins, E. M. (Eds.),
International handbook of adult mortality
(pp.
241
261
).
Dordrecht, The Netherlands
:
Springer
.
Inkeles, A. (
1996
).
Making men modern: On the causes and consequences of individual change in six developing countries
. In Inkeles, A., & Sasaki, M. (Eds.),
Comparing nations and cultures: Readings in a cross-disciplinary perspective
(pp.
571
585
).
Englewood Cliffs, NJ
:
Prentice Hall
.
Jeon, H., Salinas, D., & Baker, D. (
2015
).
Non-linear education gradient across the nutrition transition: Mothers’ overweight and the population education transition
.
Public Health Nutrition
,
18
,
3172
3182
. 10.1017/S1368980015001640
Leon, J., Baker, D., Salinas, D., & Henck, A. (Forthcoming).
Is education a risk factor or social vaccine against HIV/AIDS in sub-Saharan Africa? The effect of schooling across public health periods
.
Journal of Population Research
.
LeVine, R. A., LeVine, S. E., Rowe, M. L., & Schnell-Anzola, B. (
2004
).
Maternal literacy and health behavior: A Nepalese case study
.
Social Science & Medicine
,
58
,
863
877
. 10.1016/S0277-9536(03)00261-2
Link, B. G. (
2008
).
Epidemiological sociology and the social shaping of population health
.
Journal of Health and Social Behavior
,
49
,
367
384
. 10.1177/002214650804900401
Link, B. G., & Phelan, J. (
1995
).
Social conditions as fundamental causes of disease
.
Journal of Health and Social Behavior
,
35
(
Extra issue
),
80
94
. 10.2307/2626958
Link, B. G., & Phelan, J. C. (
1996
).
Understanding sociodemographic differences in health—The role of fundamental social causes
.
American Journal of Public Health
,
86
,
471
473
. 10.2105/AJPH.86.4.471
Link, B. G., & Phelan, J. C. (
2000
).
Evaluating the fundamental cause explanation for social disparities in health
. In Bird, C. E., Conrad, P., & Freemont, A. M. (Eds.),
The handbook of medical sociology
(5th ed., pp.
33
46
).
Upper Saddle River, NJ
:
Prentice Hall
.
Lutz, W., & Samir, K. C. (
2011
).
Global human capital: Integrating education and population
.
Science
,
333
,
587
592
. 10.1126/science.1206964
Lutz, W., Sanderson, W., & Scherbov, S. (
2001
).
The end of world population growth
.
Nature
,
412
,
543
545
. 10.1038/35087589
Marmot, M. (
2004
).
The status syndrome
.
New York, NY
:
Henry Holt
.
Martorell, R., Khan, L. K., Hughes, M. L., & Grummer-Strawn, L. M. (
2000
).
Obesity in women from developing countries
.
European Journal of Clinical Nutrition
,
54
,
247
252
. 10.1038/sj.ejcn.1600931
McLaren, L. (
2007
).
Socioeconomic status and obesity
.
Epidemiologic Reviews
,
29
,
29
48
. 10.1093/epirev/mxm001
Miech, R. (
2008
).
The formation of a socioeconomic health disparity: The case of cocaine use during the 1980s and 1990s
.
Journal of Health and Social Behavior
,
49
,
352
366
. 10.1177/002214650804900308
Miech, R., Pampel, F., Kim, J., & Rogers, R. G. (
2011
).
The enduring association between education and mortality: The role of widening and narrowing disparities
.
American Sociological Review
,
76
,
913
934
. 10.1177/0003122411411276
Mirowsky, J., & Ross, C. E. (
2003
).
Education, social status, and health
.
New York, NY
:
Walter de Gruyter, Inc.
.
Mokdad, A. H., Serdula, M. K., Dietz, W. H., Bowman, B. A., Marks, J. S., & Koplan, J. P. (
1999
).
The spread of the obesity epidemic in the United States, 1991–1998
.
JAMA
,
282
,
1519
1522
.
Montez, J. K., Hummer, R. A., Hayward, M. D., Woo, H., & Rogers, R. G. (
2011
).
Trends in the educational gradient of U.S. adult mortality from 1986 to 2006 by race, gender, and age group
.
Research on Aging
,
33
,
145
171
. 10.1177/0164027510392388
Morris, M., & Kretzschmar, M. (
1997
).
Concurrent partnerships and the spread of HIV
.
AIDS
,
11
,
641
648
. 10.1097/00002030-199705000-00012
Müller, F. H. (
1940
).
Tabakmissbrauch und lungencarcinom
[Tobacco abuse and lung carcinoma]. Zeitschrift Fur Krebsforschung/Journal of Cancer Research and Clinical Oncology,
49
,
57
85
.
Musinguzi, J., Kirungi, W., Akol, Z., Opio, A., Biryahwaho, B., & Mulumba, N. (
2010
).
The HIV/AIDS epidemiological surveillance report 2010
.
Kampala
:
Uganda Ministry of Health
.
National AIDS Control Programme
. (
2007
).
HIV/AIDS/STI surveillance report, January–December 2005
(Report No. 20).
Dar es Salaam
:
Tanzania Ministry of Health and Social Welfare
.
Pamuk, E. R., Fuchs, R., & Lutz, W. (
2011
).
Comparing relative effects of education and economic resources on infant mortality in developing countries
.
Population and Development Review
,
37
,
637
664
. 10.1111/j.1728-4457.2011.00451.x
Peters, C., VanderEnde, K., Thorpe, S., Bardin, L., Bleiberg, A., Yount, K., & Johnson, E. (
2014
,
November
).
Women’s empowerment and its relationship to current contraceptive use in low, lower-middle, and upper-middle income countries: A systematic review of the literature,
Paper presented at the 142nd annual meeting and exposition of the American Public Health Association, New Orleans, LA. Retrieved from https://apha.confex.com/apha/142am/webprogram/Paper309550.html
Peters, E., Baker, D., Deickmann, N., Leon, J., & Collins, J. (
2010
).
Explaining the education effect on health: A field study in Ghana
.
Psychological Science
,
21
,
1369
1376
.
Peters, E., Västfjäll, D., Slovic, P., Mertz, C. K., Mazzocco, K., & Dickert, S. (
2006
).
Numeracy and decision making
.
Psychological Science
,
17
,
407
413
.
Peters, J., & Büchel, C. (
2011
).
The neural mechanisms of inter-temporal decision-making: Understanding variability
.
Trends in Cognitive Sciences
,
15
,
227
239
.
Pi-Sunyer, F. X. (
1999
).
Comorbidities of overweight and obesity: Current evidence and research issues
.
Medicine & Science in Sports & Exercise
,
31
(
Suppl
),
S602
S608
. 10.1097/00005768-199911001-00019
Popkin, B. M., & Gordon-Larson, P. (
2004
).
The nutrition transition: Worldwide obesity dynamics and their determinants
.
International Journal of Obesity
,
28
(
Suppl. 3
),
S2
S9
. 10.1038/sj.ijo.0802804
Smith, W., Anderson, E., Salinas, D., Horvatek, R., & Baker, D. (
2015
).
A meta-analysis of education effects on chronic disease: The causal dynamics of the population education transition curve
.
Social Science & Medicine
,
127
,
29
40
.
Smith, W., Salinas, D., & Baker, D. P. (
2012
).
Multiple effects of education on disease: The intriguing case of HIV/AIDS in sub-Saharan Africa
. In Wiseman, A. & Glover, R. N. (Eds.),
International perspectives on education and society Vol. 18: The impact of HIV/AIDS on education worldwide
(pp.
79
104
).
Bingley, UK
:
Emerald Group Publishing Limited
.
Smith-Greenaway, E. (
2013
).
Maternal reading skills and child mortality in Nigeria: A reassessment of why education matters
.
Demography
,
50
,
1551
1561
. 10.1007/s13524-013-0209-1
Swidler, A., & Watkins, S. C. (
2007
).
Ties of dependence: AIDS and transactional sex in rural Malawi
.
Studies in Family Planning
,
38
,
147
162
. 10.1111/j.1728-4465.2007.00127.x
Thornton, A. (
2001
).
The developmental paradigm, reading history sideways, and family change
.
Demography
,
38
,
449
465
. 10.1353/dem.2001.0039
Van Hook, J., Altman, C. E., & Balistreri, K. S. (
2013
).
Global patterns in overweight among children and mothers in less developed countries
.
Public Health Nutrition
,
16
,
573
581
. 10.1017/S1368980012001164
Wamoyi, J., Fenwick, A., Urassa, M., Zaba, B., & Stones, W. (
2011
).
“Women’s bodies are shops”: Beliefs about transactional sex and implications for understanding gender power and HIV prevention in Tanzania
.
Archives of Sexual Behavior
,
40
,
5
15
. 10.1007/s10508-010-9646-8
WHO report on the global tobacco epidemic, 2011: Warning about the dangers of tobacco
. (
2011
).
Geneva, Switzerland
:
WHO
.
Global health observatory: Overweight and obesity
[Database]. (
2013
).
Geneva, Switzerland
:
WHO
.
World health statistics 2015
. (
2015
).
Geneva, Switzerland
:
WHO
.

Supplementary data